Quantum AI Based Diagnosis System for Early Brain Tumor and Lung Cancer Detection

Abstract

Early and correct diagnosis plays a key role in enhancing the outcome in brain tumor and lung cancer patients. Traditional Artificial Intelligence (AI) and Deep Learning (DL) models are handicapped by expensive computational costs and challenges in handling high-dimensional medical data. This research article presents a Quantum AI based Diagnostics System employing a hybrid quantum-classical computing model. The framework combines traditional feature preprocessing with the exponential power of quantum computing. For classification in brain tumors (e.g., LGGs/HGGs), the quantum core uses a Hybrid Quantum-Classical Integrated Neural Network (HQCINN) or Variational Quantum Classifier (VQC). For lung cancer prediction, the framework might employ quantum-enhanced clustering such as Quantum-Enhanced K-Medoids or optimization models such as Quantum --Genetic Binary Grey Wolf Optimizer (Q-GBGWO) with Extreme Learning Machines (ELM). This hybrid methodology is intended to achieve superior diagnostic efficacy and speed across both MRI and CT modalities, laying the groundwork for faster and more individualized clinical diagnostics.

Introduction

Early and accurate diagnosis of malignancies like brain tumors and lung cancer is critical for effective intervention. Classical AI and deep learning (DL) models are promising but face high computational costs and challenges in handling high-dimensional medical data. Quantum AI leverages phenomena like superposition and entanglement to enhance processing power and efficiency.

Key Quantum AI Models & Techniques

1. Brain Tumor Classification

• Hybrid Quantum-Classical Integrated Neural Network (HQCINN) and Variational Quantum Classifier (VQC) are used for multi-class tumor classification.

• Data preprocessing: MRI images reshaped and normalized (Min-Max normalization).

• Quantum layer: Parameterized Quantum Circuit (PQC) transforms input data into Hilbert space for advanced feature analysis.

• Output: Expectation values from the quantum circuit are fed into a classical Softmax layer.

• Training: Hybrid iterative optimization using Gradient Descent or Analytic Quantum Gradient Descent (AQCD) to minimize cross-entropy loss.

2. Lung Cancer Prediction

• Quantum-Enhanced K-Medoids Clustering improves classical clustering using Manhattan Distance in a quantum-enhanced high-dimensional space.

• Quantum-Genetic Binary Grey Wolf Optimizer (Q-GBGWO) with Extreme Learning Machine (ELM) optimizes ELM input weights for higher classification accuracy.

• Preprocessing: Classical feature normalization ensures effective quantum encoding.

• Quantum-enhanced computations improve cluster selection, resulting in more accurate lung cancer predictions.

Algorithms & Workflows

• Q-GBGWO-ELM: Combines quantum optimization with ELM for multi-cancer detection; uses hybrid feature fusion for improved imaging feature extraction.

• VQC for Glioma Classification: Classical ensemble feature selection identifies molecular and clinical markers; PQC encodes features; expectation values guide loss minimization.

• HQCINN for Multi-Class Brain Tumors: Classical preprocessing → quantum feature transformation → classical output; iterative hybrid learning refines classification.

• Quantum-Enhanced K-Medoids: Improves distance calculations using quantum computation, enhancing lung cancer clustering and prediction accuracy.

Advantages

• Higher accuracy in tumor and cancer classification.

• Faster training and inference compared to classical-only AI models.

• Efficient handling of high-dimensional data.

• Hybrid quantum-classical approach balances computational efficiency and practical feasibility.

Conclusion

The four experiments together demonstrate the substantial potential of quantum-enhanced Artificial Intelligence (AI) on a wide range of cancer diagnostic problems, both brain and lung cancer. Hybrid Quantum - Classical Integrated Neural Network (HQCINN) and Variational Quantum Classifier (VQC) models utilise the computational strength of Parameterised Quantum Circuits (PQCs) and quantum features such as entanglement for classification, especially in very-high-dimensional feature spaces such as those from MRI images and molecular markers. In parallel, the Quantum - Genetic Binary Grey Wolf Optimiser - Extreme Learning Machine (Q-GBGWO-ELM) solves the issue of multi-cancer detection by merging deep traditional feature extraction with a quantum-optimised, rapid learning algorithm. Independently, the Quantum-Enhanced K-Medoids model targets the problem of unsupervised learning, enhancing patient clustering through incorporating quantum principles in order to optimise the calculation of the distance metric. The Q-GBGWO-ELM model, which aims for early multi-cancer detection, is perhaps the best overall model considering its wider scope and purported efficiency on a very complex, multi-modal dataset. The model\'s power is in its overall architecture that combines FuNet transfer learning for enhanced multi-modal feature fusion with an Extreme Learning Machine (ELM) whose parameters are optimised through the Quantum-Genetic Binary Grey Wolf Optimiser. This architecture enables it to address the most ambitious and most broadly relevant challenge—detection of multiple cancer types from different data sources—and results show it enhances diagnostic accuracy by an average of 6% compared to other models tested, illustrating its strength and better performance in a heterogeneous diagnostic setting. The remaining three quantum-augmented strategies would be the best option in certain, targeted clinical situations wherein they provide specialised benefits in comparison to the general multi-cancer classifier. The HQCINN model suits complex, multi-class image classification tasks, for example, differentiating various forms of brain tumours (e.g., meningioma, glioma) directly from MRI scans, where its hybrid architecture, based on shallow or deep circuits, is tuned to work on and distil high-level features from image data. The VQC is the better option for binary classification of highly discriminative molecular data, e.g., predicting LGGs vs. HGGs prognosis risk with a small, feature-reduced subset of TCGA clinical and genetic markers with high metrics, such as an AUC of 0.94, that are essential for accurate risk stratification. Lastly, the Quantum-Enhanced K-Medoids model is best for unsupervised patient segmentation or identification of inherent groupings within a dataset, as opposed to direct prediction, due to the fact that it employs quantum principles to optimise the distance metric and form stronger clusters between patients for subgroup investigation or treatment customisation.

References

[1] Bilal, A., Shafiq, M., Obidallah, W. J., Alduraywish, Y. A., & Long, H. (2025). Quantum computational infusion in extreme learning machines for early multi-cancer detection. Journal of Big Data, 12(1), 27. [2] Akpinar, E., & Oduncuoglu, M. (2025). Hybrid classical and quantum computing for enhanced glioma tumour classification using TCGA data. Scientific Reports, 15(1), 25935. [3] Akpinar, E., Islam, S. M., & Oduncuoglu, M. (2025). Multi-classification of brain tumours using proposed hybrid quantum–classical integrated neural network (HQCINN) models: shallow and deep circuit approaches. Neural Computing and Applications, 1-32. [4] Solikhun, S., Pujiastuti, L., & Wahyudi, M. (2025). Enhancing Lung Cancer Prediction Accuracy Using Quantum-Enhanced K-Medoids with Manhattan Distance. MATRIK: Jurnal Manajemen, Teknik Informatika dan Rekayasa Komputer, 24(3), 493-506.

Copyright

Copyright © 2025 Dr. S. Gunasekaran, Abhiram S. Nair, Adharsh N, Ajay Shankar, Midhun M. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.